Queue consumer invocations failing as UNHANDLED_EXCEPTION with no app logs since 2026-10-19 07:34 UTC

Hi all,

Since this morning (2026-10-19, ~07:34 UTC), about 20% of invocations of our async events consumer fail as UNHANDLED_EXCEPTION in the developer console metrics. That keeps tripping the “Invocation success rate” alert. Nothing changed on our side: the same code ran from Oct 15 with zero errors.

Setup:

  • Web trigger pushes Linear webhook events to a queue (queue.push with concurrency: { key, limit: 1 }), and a consumer module processes them (timeoutSeconds 120)
  • App ID: ec111fc8-df5d-4a5f-b0cc-d5eba3b2959d, production environment

What we see:

  • Only the consumer function errors. The web trigger, product event triggers and resolver are all clean.
  • The failing invocations write nothing to the app logs. No error, no stack.
  • Every event we pushed is still processed exactly once and logged normally. Nothing comes back with a retryContext, so we’re not losing data.
  • We deployed a version that wraps the entire consumer handler in try/catch and logs any throw. The catch never fires, and the error rate is unchanged (~22% over the last 3 hours).
  • We also checked the whole codebase for floating promises (typescript-eslint no-floating-promises): zero.
  • Possibly related: three forge deploy runs this morning failed at “Deploying to environment” with a generic “Deployment failed” and succeeded on a plain re-run.

So it looks like the exception is raised outside our handler, in the runtime or queue layer.

Is anyone else seeing this on their queue consumers today? And could someone from Atlassian take a look? We’ve opened a support ticket too: Jira Service Management

Thanks!

Update in case someone else hits this: we found the trigger.

It was timeoutSeconds: 120 on the consumer function. With that set, Forge produced extra consumer invocations that failed as UNHANDLED_EXCEPTION before our handler ever ran. There was no app log, about 4 per hour regardless of traffic, and dev and prod started failing in the same minute.

How we confirmed it: our dev and prod installs receive the same webhook traffic, so we ran dev on the default timeout and left prod on 120s.

  • Over 9 hours, dev had 0 errors in 38 invocations.
  • Prod had 37 errors in 77 invocations (77 ≈ the same 38 real runs plus 37 phantom ones).
  • After shipping the default timeout to prod, errors dropped to 0 immediately.

Our consumer runs take ~0.25s (p99 under 2s), so the 120s was only headroom we never needed.

Things that did NOT help, in case it saves someone time:

  • wrapping the whole handler in try/catch (it never fired)
  • removing the concurrency option from queue.push

Atlassian folks: it still looks like a platform bug to me. A queue consumer with a raised timeoutSeconds shouldn’t get phantom failing invocations. Happy to share the app ID, environment, invocation timestamps or metrics if it helps.